Voluntary DBA

When we manage databases in either the FULL or BULK_LOGGED recovery models, we need to pay particular attention to the size of the transaction log files. If our processes aren’t optimal, we can see log files grow either out of control, or beyond what we think is a reasonable size.

Virtual Log Files

As I mentioned in a previous post, the transaction log contains an ordered sequence of the physical operations that occur on a database.

What I didn’t mention is that the physical transaction log file also contains logical sections, called virtual log files (VLFs). It’s the VLFs which actually contain the physical operations I mentioned. The purpose of VLFs is to allow SQL Server to more efficiently manage the log records; specifically, to know which portions of the transaction log are used, and which aren’t.

Knowing the portions of the log that are used is important when we go to take a transaction log backup, which creates a copy of all the transaction log records since the last transaction log backup. After a transaction log backup, as long as all the log records within a VLF are not part of an active transaction (or are required for some other feature, such as replication), the VLF can be marked as unused/inactive. This allows SQL Server to reuse that portion of the log file. This process is called many names, including clearing the log and log truncation. It does not affect the physical size of the log file.

Problems only start happening if all the VLFs end up being used, and none are available¹. This means the log file (the physical file) has to grow to accommodate more transactions. A physical log file growth automatically creates new unused VLFs, and so the transactions can continue.

What is causing the log to grow after I shrank it?

Any write operation in the database can potentially cause the log file to grow. The real question is: why are all the VLFs used up when that’s not what is expected?

Here are some of the problems you might encounter:

• Not taking transaction log backups frequently enough, or not taking them at all. The only way to mark VLFs as inactive is to take a transaction log backup (again, only for FULL and BULK_LOGGED databases). If transaction log backups are only taken once/day, besides exposing yourself to a full day of data loss, you also need a transaction log file large enough to hold all the transactions that occur for a full day. That could be quite a bit of space! The solution is obvious, and it’s a win for recoverability, too: take transaction log backups more frequently.
• Overly aggressive and/or too frequent index maintenance. I’m certainly a proponent of index maintenance, but it’s very easy to get carried away. Maintenance plans that rebuild shiny new indexes for you every night are both ridiculously easy to set up, and also ridiculously bad for the size of the transaction log. Typically, rebuilding all the indexes of a database takes more log space than the data files actually take up. If you’re dealing with databases of even small-to-moderate sizes (say, 1 GB and larger), this can add up to a tremendous amount of wasted storage space, because all that transaction log has to be backed up at some point, and the log file will likely end up larger than the data files. What I strongly recommend doing is putting in place a much more selective index maintenance process, such as Ola Hallengren’s index maintenance scripts, which I use in production.
• The database is in FULL or BULK_LOGGED when it actually should be in SIMPLE. Before you go flipping the settings around (which can potentially cause data loss), go read about choosing the right recovery model for the database. Typically, I see this happen when a production backup is restored to a development environment, and a backup gets taken for some reason. And then the transaction log ends up growing, and growing, and growing as developers test and roll back their work, right up until things screech to a halt when the drive that contains the log file has filled up.
• A SQL Server feature, such as replication or mirroring, is enabled. Until the transactions have been copied to the secondary server, the records cannot be cleared from the transaction log. If the secondary is slow applying changes, or if you don’t even have a secondary any more and things weren’t cleaned up correctly, this could lead to infinite transaction log growth. To solve this really depends on where the specific problem is — solve that first, and then see if there is still an issue with the transaction log.

If none of those cover your scenario, you can run the following query, which will tell you why the transaction log is still in use.

SELECT
    name,
    log_reuse_wait_desc,
    recovery_model_desc
    FROM sys.databases

 
¹ Extra empty log space is pre-allocated such that there is always enough space for active transactions to be able to roll back. Regardless, that space is “used” in the sense that it’s been spoken for already.